High-Precision Prioritization for Arabic Bug Reports: Fine-Tuning asafayaBERT with Transfer Learning
Batool Armouti, Sara Khader, Abdallah D. Qusef, Ahmad Althunibat · Procedia Computer Science · 2026
Prioritizing software bugs is critical for development teams; to efficiently allocate resources and address severe issues promptly. While extensively studied for English reports, Arabic bug prioritization remains largely unexplored despite growing Arabic software usage. This research bridges the gap by fine-tuning pre-trained Arabic language models asafayaBERT and araGPT-2. asafayaBERT is a pre-trained CNN with BERT model, fine-tuned for classification of Arabic language. The two models are fed with Arabic bug priority dataset to classify their priority levels. Experimental results demonstrate that asafayaBERT achieves superior performance with 76.63% precision araGPT-2. Performance in deployment environment is an important factor for such applications. asafayaBERT shows more stable performance in addition to the significant lower training time than araGPT-2. The findings confirm that automated prioritization of Arabic bug reports is feasible and can achieve practical performance levels. This approach can significantly reduce manual triage effort for Arabic-speaking development teams, saving valuable time in software maintenance while improving response times for critical bugs.